upload_files()/upload_files_async() raises httpx.RequestNotRead

glean/api_client/_hooks/init.py の XGlean.before_request() から始め、次に examples/agent_with_file_upload.py を確認し、ドキュメント化された入力で失敗するアップロードを再現します。upload_files() と upload_files_async() が一覧にあるコンテンツのバリエーションで動作し、非 multipart の JSON エンドポイントに影響がないことが完了の条件です。

Description
The official example examples/agent_with_file_upload.py fails with httpx.RequestNotRead when calling glean.client.chat.upload_files().

Steps to reproduce
Run the official example as-is:

from glean.api_client import Glean, models
with Glean(api_token="...", instance="...") as glean:
upload_result = glean.client.chat.upload_files(
files=[models.File(file_name="test.csv", content=b"a,b\n1,2")]
)
All content variants produce the same error: bytes, io.BytesIO, open(..., 'rb'), FileTypedDict.

Error
httpx.RequestNotRead: Attempted to access streaming request content, without having called read().

Root cause
The XGlean BeforeRequestHook always reconstructs the httpx.Request object to add optional headers:

glean/api_client/_hooks/init.py — XGlean.before_request()

return httpx.Request(
method=request.method,
url=request.url,
headers=new_headers,
content=request.content, # <-- raises RequestNotRead
extensions=request.extensions,
)
For multipart requests (file uploads), httpx builds the body as a stream. Accessing request.content before request.read() raises httpx.RequestNotRead. This happens on every upload_files() / upload_files_async() call, regardless of how file content is passed, because the hook runs before the request is sent.

Non-multipart endpoints (JSON body) are unaffected because request.content is available immediately.